Effects of Uninsurance during the Preceding 10 Years
on Health Status among Rural Working Age Adults
1
Effects of Uninsurance during the Preceding 10 Years
on Health Status among Rural Working Age Adults
Authors:
Janice C. Probst, PhD Charity G. Moore, MSPH, PhD
M. Paige Powell, PhD William Pearson, PhD
Amy Brock Martin, PhD
South Carolina Rural Health Research Center 220 Stoneridge Drive, Suite 204
Columbia, SC 29210 (803) 251-6317
Janice C. Probst, PhD, Director
September 2005
Funding acknowledgement:
This report was prepared under Grant No. 6 U1C RH 00045-03
Office of Rural Health Policy Health Resources and Services Administration US Department of Health and Human Services
Rockville, Maryland Joan Van Nostrand, DPA, Project Officer
i
Executive Summary
Our study sought to determine if individuals with longer periods of uninsurance, in multivariable analyses controlling for income, poverty and health status/behavior at the beginning of the time period, will be more likely to be overweight, to report experiencing hypertension or diabetes, or to describe their health as “fair” or “poor.” We also looked to see if the effects of uninsurance would be greater in rural than in urban respondents, and greater for minority rural populations than for white rural populations.
Key Findings Effects of Uninsurance
• In a population just reaching age 40, continuous health insurance coverage across the preceding 8 to 10 years was not associated with better self-perceived health, as measured using the SF-12 Physical Component Score, than interrupted coverage. A positive relationship between insurance coverage and health in unadjusted analysis, which remained after controlling for race and residence, ceased to be significant after age, education, marital status, and poverty during youth were held constant.
• Similar results were found when health status was dichotomized into “fair to poor” versus better self-perceived health.
• Health insurance was not significantly related to the probability that a respondent would report a high BMI (obesity), or would have been diagnosed with hypertension or diabetes
• Continuous insurance coverage was significantly related to better mental health, as measured by the SF-12 Mental Component Score. The effect persisted in multivariable analysis controlling for residence, race, and demographic characteristics.
Effects of Rural Residence and Race
• Rural residents reaching age 40 in 1998 or 2000 were less likely to have been continuously insured between 1989 – 2000 than were their urban peers. Among whites, 57.2% of rural residents were continuously insured versus 66.4% of urban residents (p=0.0009). Similar but non-significant trends were found among African Americans (37.0% versus 44.7%; p=0.1098) and among Hispanics (37.7% versus 44.1%; p = .4114).
• With differences in insurance coverage, residence, education, marital status, age, and poverty in young adulthood held constant, African Americans and Hispanics in the study did not perceive themselves to have poorer physical or mental health than whites. The effects of lack of insurance were not greater in rural than in urban areas and were not greater for minority rural populations than for white rural populations (no significant interactions).
• The odds that an individual would experience obesity, and hypertension and diabetes varied by race. Obesity and diabetes were more common among both African Americans and Hispanics at age 40 than among whites. Diabetes was more likely to be reported by African American respondents (Table 10).
• Over the 1989 – 1998/2000 period, slightly more than a third of rural minority respondents (37%) were continuously insured. While rural whites were more likely to be continuously insured (57%) than rural minorities, 43% of rural whites experienced
ii
periods without insurance. Rural whites were significantly less likely to report continuous insurance than were urban whites.
Policy Recommendations • The Department of Health and Human Services should evaluate how its component
agencies can best support local programs working to improve access to care for uninsured populations. Working with organizations such as the American College of Healthcare Executives, the American Hospital Association, and the National Rural Health Association, DHHS should ensure that best practices among such local efforts are identified and shared. Technical assistance materials developed to document best practices should be available to, and understandable by, non-specialist community leaders.
• The Secretary of the Department of Health and Human Services should work with the National Association of Rural Health Clinics, the National Rural Health Association, and the National Association of Community Health Centers to define model approaches to involving RHCs in the care of uninsured persons. It would be particularly valuable to document collaborations among RHCs, CHCs, and other service providers, and to communicate the information broadly to state and local planners.
• In partnership with local mental health professional organizations, Area Health Education Consortia are encouraged to develop strong educational programs in mental health screening, treatment, and referral for family medicine, internal medicine, and general surgery resident physicians, and for nurses. Such educational content is needed both during practitioner training and as part of continuing medical education programs.
• State Area Health Education Consortiums are encouraged to partner with local middle and high schools in an attempt to promote health careers, both as a means of encouraging children to higher educational attainment and as a way of growing a local health labor force. Such activities should be particularly promoted in rural areas.
• State Area Health Education Consortiums are encouraged to partner with agencies in their state responsible for implementation of the WIA, in order to identify health careers job training opportunities that may benefit rural residents and institutions.
Research Recommendations • The youngest members of the NLSY-79 cohort reach age 40 in 2005. When public use
data are available for the full NLSY-79 cohort, the analysis reported here should be repeated. With larger sample sizes, many limitations of the present analysis may be obviated. For example, more detailed definitions of “rural” and “insured” will be possible.
• Poverty at ages 19 – 21 appears to have lasting effects on perceived mental and physical health. This long-term effect may result from culture as learned in youth, from employment choices, or from other factors not included in the present analysis. Identification of factors which ameliorate the effect of early poverty, which may include education, employment in varying occupations, and so on, is essential.
iii
Table of Contents
Chapter One: Introduction 1 Background 1 Objectives of our research 2 Outcome measures 2 Report Format 4 Chapter Two: Results 6 Population Studied: A National Sample of “Youth” in 1979 6 Levels of Health Insurance Coverage Across Time and Health Outcomes 9
Health Insurance Coverage and Health Outcomes by Race/Ethnicity 11 Health Outcomes by Residence and Health Insurance Coverage 13 Holding Personal Characteristics Equal: Multivariable Analysis 14
Chapter Three: Conclusions and Policy Recommendations 19 Conclusions 19 Limitations to the Present Analysis 24 Policy Recommendations 26 Research Recommendations 31 Appendix A: Methods 33 The National Longitudinal Survey of Youth 33 Insurance Status 34 Residence 35 Health Status 35 Other Determinants of Health Outcomes 36 Statistical Methods 37 Appendix B: Tables 38 Appendix C: References 49
iv
List of Figures
Figure 1. Residence as Percent of Years Since 1989 Lived in SMSA by Race/Ethnicity 8 Figure 2. Health Insurance Coverage between 1989-1998/2000 by Race/Ethnicity 9 Figure 3. Physical and Mental Health Scores by Health Insurance Coverage. 10 Figure 4. Percent of Respondents Reporting Fair to Poor Physical Health, by Health Insurance Coverage. 11 Figure 5. Physical and Mental Health by Race/Ethnicity and Health Insurance Coverage. 12 Figure 6. Self-Reported Fair to Poor Health Status at Age 40, by Race/ethnicity and Health Insurance Coverage. 12 Figure 7. SF-12 Physical and Mental Component Scores at Age 40, by Residence and Health Insurance Coverage. 13
List of Tables Table B-1. Description of Population, Health Interview at Age 40 38 Table B-2. Distribution of health insurance coverage by race, residence (1979 and current), and residence over time. 39 Table B-3. Outcomes by health insurance coverage 40 Table B-4. Health Insurance and Health Outcomes Stratified by Race 41 Table B-5. Health Insurance and Health Outcomes Stratified by Residence 42 Table B-6. General Linear Model Results for SF-12 Physical Component Scores. 43 Table B-7. General Linear Model Results for SF-12 Mental Component Score 44 Table B-8. Logistic Regression Results for Factors Associated with Self-reported Fair or Poor Health Status 45 Table B-9. Logistic Regression Model Results for Predicting Obesity 46 Table B-10. Logistic Regression Model Results for Predicting Diagnosis of Hypertension 47 Table B-11. Logistic Regression Model Results for Predicting Diagnosis of Diabetes 48
v
Chapter One
Introduction
Background
A fundamental assumption in health services research is that being uninsured leads to
poorer health status due to a lack of access to care (Davis, 1997). Losing insurance, whether it is
Medicaid or private insurance, leads to an increase in the proportion of persons who report no
usual source of care, an increase in the proportion who report barriers to accessing care, and an
increase in the proportion who report dissatisfaction with their ability to obtain needed care
compared to those who kept their insurance or gained insurance (Kasper, Giovannini, and
Hoffman, 2000). Similarly, persons whose insurance coverage is interrupted, report significant
access barriers (Schoen, DesRoches, 2000). Uninsured adults make fewer health care visits than
their insured counterparts (Mueller, Patil and Ullrich, 1997; Comer, Mueller, Blankenau, 2000;
Ayanian, Weissman, et al, 2000). Reduced visit frequency, in turn, has been associated with
poorer outcomes for hypertension and vision correction (Keeler, Brook et al, 1985; Lurie,
Kambert et al, 1989). However, lower visit frequency associated with high coinsurance
requirements was not associated with other health outcomes in the Rand Health Insurance
Experiment, which tracked differently insured populations for three to five years (Keeler, Brook
et al, 1987).
Rural populations are known to have lower rates of health insurance, with rural minorities
being more likely to lack insurance than either rural whites or urban minority populations
(Glover, Moore et el 2004; Probst, Moore, in press). Uninsured rural residents, in turn, are
significantly less likely to report a health care visit during the past year than their peers, even
controlling for factors such as self-perceived health status, education, and employment (Glover,
1
Moore et al 2004). The long term effects of rural residence and insurance, however, are not yet
known.
Objectives of Our Research
We sought to examine the effects of lack of health insurance and rural residence
simultaneously over time, using a unique long-term study, the National Longitudinal Survey of
Youth 1979 (NLSY). The NLSY, as its name implies, originated as a study of young people
(ages 14 – 21). The members of the original 1979 panel have been surveyed at regular intervals
for the subsequent 25 years. Beginning in 1989, the NLSY included questions regarding health
insurance coverage. When respondents turn 40, the survey added a series of health-related
questions. Working with the publicly available data sets from each year of the NLSY, we are
able to examine how health insurance status and rural residence (defined as living in a non-
metropolitan county) work jointly to influence health status indicators in early middle age.
The study had two objectives:
• To ascertain whether periods of uninsurance over time, in multivariable analyses
controlling for income, poverty and health status at the beginning of the time period,
lead to poorer health outcomes at age 40. The outcomes studied include self-reported
mental and physical health status, body mass index (BMI), and diagnosed disease
(hypertension, diabetes).
• To ascertain whether the effects of uninsurance are greater for rural than for urban
respondents, and greater for minority rural populations than for white rural
populations.
Outcome measures
Health Status: Because of the difficulty and prohibitive cost of conducting physical
examinations for survey research, much research depends on individual’s reports of their own
health. We use two self-report measures for physical health: a general self-report of health, and
2
the SF-12 Physical Component Score. Previous studies of the effects of insurance, both cross-
sectional and longitudinal, have used such subjective measures of health status. Some cross
sectional research has found the subjective health status of adults without insurance to be lower
than the subjective health status of those with insurance (Franks, Clancy, et al, 1993). Type of
insurance also affects self-perceived health; persons receiving public insurance often report
poorer health status than either privately insured or uninsured persons (e.g., Ross and Mirowsky,
2000). However, longitudinal studies of the relationship between insurance coverage and health
offer conflicting results. Over a two-year period, working age adults who lost health insurance
coverage or had gaps in coverage, compared to their continuously insured peers, were not more
likely to experience an adverse event due to health (Kasper et al, 2000) or to perceive their health
status as fair to poor (Schoen, DesRoches, 2000; Kasper et al, 2000). On the other hand, Baker,
Sudano and colleagues, in several analyses of adults ages 51 to 61 across a four-year period
(1992 –1996), found poorer health outcomes among continuously uninsured and intermittently
uninsured than among continuously insured (Baker, et al., 2001, 2001, 2002). Differences in
findings between the two sets of studies may stem from differences in the time period (two
versus four years) or in the ages of the study participants.
Rural working age adults report poorer health status than their urban counterparts
(Glover, Moore et al, 2004). The effects of rural residence when insurance status over time and
other variables are held equal has not yet been examined.
Mental health status: Accurate clinical assessment of the prevalence of mental disorders
would be as prohibitively expensive as clinical assessment of physical status. Thus, we again
rely on a standardized self-report proxy, the SF-12 Mental Component Score. While the Rand
Health Insurance Experiment found no variation in self-reported mental health status among
3
insured persons with high coinsurance requirements (Wells, Manning, Valdez, 1989), that
research did not address the situation of completely or intermittently uninsured persons. More
recent work has linked health insurance to mental health status, although only among single
persons (Ferrer, Palmer, Burge, 2005).
Chronic disease and risk factors: hypertension, diabetes, and obesity. The NLSY
inquired about a variety of health conditions, such as arthritis, diabetes, hypertension, cancer, and
other diagnoses. In the age-40 NLSY population, most conditions were rare. The number of
persons who had developed hypertension or diabetes, however, was large enough that the effects
of rural residence and insurance on the prevalence of these two conditions could be studied.
Hypertension and diabetes are more common among minorities, including rural minorities
(Mainous, King et al 2004). Both conditions are associated with physical inactivity and obesity;
inactivity and obesity, in turn, are more common in rural areas and among rural minorities in
particular (Patterson, Moore et al, 2003). The rationale for studying these conditions, beyond
their prevalence, is the possibility that appropriate access to care would expose an individual to
more frequent counseling about diet, exercise and weight, thus lowering the prevalence of
obesity and its potential sequellae, hypertension and diabetes (USPTS, 2003). Among
individuals in late middle age, uninsured and intermittently insured persons have been shown to
reduce their use of health care (Johnson and Crystal, 2000), particularly preventive services
(Sudano and Baker, 2003). Reduced use of care might eventually result, we hypothesized, in an
increased prevalence of disease.
Report Format
The principal results of the study are presented in Chapter Two, with conclusions and
policy implications offered in Chapter Three. Details about the National Longitudinal Survey
4
of Youth – 1979 and about the methods used for this report are presented in Appendix A, while
the detailed tables of analytic results are provided in Appendix B.
5
Chapter Two
Results
Population Studied: A National Sample of “Youth” in 1979
We analyzed data from the 1998 and 2000 administrations of the National Longitudinal
Survey of Youth--1979, limiting the analysis to persons who turned 40 in one of those years. At
age 40, survey participants were asked a variety of questions pertaining to their physical and
mental health. A total of 3,164 respondents, aged 19 through 21 when originally contacted by
the NLSY in 1979, reached age 40 by the 1998 or 2000 survey. Among these, 3,104 had
sufficiently complete residence and insurance information each year to be included in the current
analysis. A more detailed discussion of the dataset and methods can be found in Appendix A.
All Tables are located in Appendix B.
About half of the respondents were white, nearly a third were Hispanic and
approximately one-fifth were African American (Table B-1). The study group contained slightly
more women than men, a pattern consistent across race/ethnicity groups. Minority groups
differed from white respondents in several characteristics linked with health. African Americans
were about twice as likely and Hispanics nearly three times as likely as whites to have less than a
high school education. However, the proportion of persons with a college degree or better did not
differ across race/ethnicity. Whites were markedly more likely to be married (72.5%) than were
African Americans (36.5%); Hispanics were similar to whites (60.9% currently married).
The National Longitudinal Survey of Youth (NLSY-79), commissioned by the Bureau of Labor
Statistics, is one of the nation’s longest running panel surveys (NLSY79, 2001). This multistage, stratified
clustered sample was initiated in 1979. Participants were interviewed annually from 1979 through 1994,
and biannually until present, with an 84% retention rate through 1998.
6
In 1979, 38.9% of African American youths surveyed by the NLSY, and 28.8% of
Hispanic youths, were in poverty, compared to only 9.9% of whites. By comparison, the 1980
Census found 11.4% of white children aged 18 and below to be living in poverty in 1979, as well
as 40.7% of African American children and 27.4% of Hispanic children (Bureau of the Census,
1981). Differences between the Census population and that studied by the NLSY are probably
attributable to age differences, as the NLSY respondents were aged 14 – 21 in 1979. Decreases
in the proportion of respondents living in poverty between initial 1979 interviews and age 40
interviews are worth noting. The proportion of Hispanics living in poverty decreased by 47%,
African Americans by 36%, and whites by 24% (Table B-1). Nonetheless, given the differences
in starting proportions, poverty continued to be more prevalent at age 40 among minority adults
(24.8% among African Americans, 15.2% among Hispanics) than among whites (7.5%).
Persons were more likely to remain poor between 1979 and 1998/2000 than to become poor.
Among persons who were poor in 1979, 22.7% remained poor when interviewed at age 40, while
among persons not in poverty in 1979, only 7.7% were poor at age 40 (p=0.0000).
Key Measure: Rural Residence
Residence changed over time, making identification of a “rural” population problematic.
In 1979, about a third of white youths (34.6%), one-quarter (25.9%) of African American youths
and 16.1% of Hispanic youths surveyed lived outside of metropolitan areas. By age 40,
however, the proportion of individuals currently living in non-metropolitan areas had dropped to
9.8% for whites, 5.2% for African Americans, and 3.5% for Hispanics. Since our insurance
measure only covered the period 1989 through 2000, for consistency in exposure variables, we
defined residence using the same time period. To characterize persons as “urban” or “rural,” we
used the proportion of time each respondent lived outside a metropolitan area during the 1989 –
7
1998/2000 time period.1 As shown in Figure 1, most respondents resided most of the time in
urban areas. When characterizing respondents by residence, we defined them as “rural” if they
had lived more than half of the years between 1989 and the time at which their health was
measured (age 40) outside metropolitan areas. About a quarter of white respondents (24.8%)
lived at least half of the time between 1989 and their 40th birthday in non-metropolitan areas, as
did 14.0% of African American respondents and 8.1% of Hispanic respondents (See Table B-1).
Figure 1. Residence as Percent of Years Since 1989 Lived in SMSA by Race/Ethnicity
8.1 14.024.8
92.9 86.075.2
0
20
40
60
80
100
Hispanic African American White
0-49% in SMSA 50-100% in SMSA
Key Measure: Health Insurance Coverage
Because health insurance was examined over approximately 12 years, the proportion of
time that a respondent had coverage was used as the insurance measure. Insurance was defined
as any coverage, whether public or private.2 About 54% of the study population was
1 To test the sensitivity of our analysis to the time period used, we also characterized residence based on the entire 1979 – 1998/2000 time frame. Results were not appreciably different; see methods section. 2 The category “insured” intentionally combines private and public insurance coverage, with no attempt to distinguish between the two. Adults in the age group studied here can only obtain public insurance, principally Medicaid, for childbirth related expenses or for disability. Because Medicaid disability coverage would be provided based on poor physical or mental health, it would be too logically entwined with the study’s outcomes to be fruitful analytically. Inclusion of persons who might have received Medicaid in the “insured” group biases the study in a conservative direction, that is, makes it less likely to find that insurance coverage is associated with better health.
8
continuously covered throughout the years when the NLSY included questions about health
insurance (1989 to 2000; Table B-2). An additional 14.8% of respondents reported being covered
between 75% and 99% of the time. Only 16.7% were uninsured for more than half of the time.
Insurance coverage varied with race/ethnicity (Table B-2 and Figure 2). Whites were
more likely to be continuously covered than African Americans or Hispanics. Coverage was not
significantly associated with residence in 1979 or residence at age 40. However, whites living
principally in urban areas during the 1989 – 1998/2000 period were significantly more likely to
have been continuously insured than their rural peers (Figure 2). Differences were similar in
direction for minority populations, although not statistically significant.
Figure 2. Health Insurance Coverage between 1989-1998/2000 by Race/Ethnicity
37.7 37.0
57.2
44.1 44.7
66.4
01020304050607080
Hispanic African American WhiteRural³ 50% Urban
P values for urban/rural comparison: Hispanic, p = 0.4113; Afr Amer, p = 0.1098; white, p = .00098
Levels of Health Insurance Coverage Across Time and Health Outcomes
To assess overall health status, the NLSY administered two commonly used instruments,
the SF-12 Physical Component and the SF-12 Mental Component. These scales yield an average
value of 50 in a standard population; higher values are associated with better physical or mental
health (Ware, Kosinski, and Keller, 1995). In general, NLSY respondents surveyed at age 40
perceived themselves to be in good physical and mental health; mean values are above 50 on
both scales across all categories of insurance coverage.
9
Without controlling for factors such as race, residence, or other personal characteristics,
the proportion of time with insurance coverage had significant effects on both physical (p =
0.0004) and mental (p = 0.0001) health status (See Figure 3 and Table B-3). Persons who were
continuously covered had the highest scores for both physical and mental health at age 40. For
physical health, there is a possible threshold effect, with persons lacking continuous insurance
having lower scores than persons who were consistently insured. For mental health, the
relationship appears to be U-shaped.
Figure 3. Physical and Mental Health Scores by Health Insurance Coverage.
53.1 53.6
51.752.6
51.5 5151.2 51.851.1
52.9
47
49
51
53
55
57
SF-12 Physical SF-12 Mental
Continuously Covered 75-99% 50-74% 25-49% 0-24%
The NLSY also assessed health status by asking persons to rate their own health status.
Insurance coverage was positively related to good health as measured by the SF-12 Physical
Component; conversely, it was negatively related to self-described ‘fair to poor” health (Figure
4, top of next page, and Table B-3). There appears to be a threshold effect, with persons who
were continuously insured or insured 75% of the time or more being less likely to report poor
health.
No clear relationship emerged between the proportion of time a respondent had insurance
coverage and the prevalence of self reported overweight or obesity (BMI calculated from
reported height and weight), or diagnosed diabetes or hypertension. While insurance coverage
10
Figure 4. Percent of Respondents Reporting Fair to Poor Physical Health, by Health Insurance Coverage.
8.313.5 16.1 16.7 17
05
1015202530
Fair to Poor Health
%
Continuously Covered 75-99% 50-74% 25-49% 0-24%
had a strong statistical relationship to self-perceived health status, no differences in diagnosed
illness or obesity were statistically significant at the p = 0.05 level.
Health Insurance Coverage and Health Outcomes by Race/Ethnicity
To examine the effects of health insurance within race/ethnicity groups, we dichotomized
the insurance measure: continuous coverage over the study period, versus coverage not
continuous over the study period. Narrower insurance categories did not contain sufficient
numbers of respondents of different race/ethnicity for valid estimates. Results of the analysis are
shown in Figure 4 and in Table B-4.
Among Hispanics and Whites, the mean for SF-12 Physical Component was higher for
persons who were continuously covered than for those who were not; differences were not
significant for African Americans (Table B-4). Among whites and African Americans, the mean
SF-12 Mental Component was significantly higher for those continuously covered compared to
those not continuously covered; differences were not significant for Hispanics (Figure 5). The
proportion of persons reporting “fair to poor” health was significantly higher among people not
continuously covered across all race/ethnicity categories (Figure 6).
11
Figure 5. Physical and Mental Health by Race/Ethnicity and Health Insurance Coverage.
53.252.1
52.753.7
53.1 53.6
51.4 51.3 50.951.9 51.9
52.7
46
48
50
52
54
56
58
SF-12 Physical(White)**
SF-12 Physical(AA)
SF-12 Physical(Hispanic)*
SF-12 Mental(White)**
SF-12 Mental(AA)*
SF-12 Mental(Hispanic)
Continuously Covered Not continously covered * p > .05 ** p > .001
Figure 6. Self-Reported Fair to Poor Health Status at Age 40,
by Race/ethnicity and Health Insurance Coverage.
7.4
13.5 12.214.3
18.3 19.1
0
5
10
15
20
25
30
% Fair to Poor (White) ** % Fair to Poor (AA) % Fair to Poor (Hisp) **Continuously Covered Not continously covered ** p > .001
Average BMI and the proportion of persons who were overweight or obese differed
across race/ethnicity groups, but did not differ by coverage status within any race/ethnicity
category. Thus, whites had the lowest average BMI and the smallest proportion of persons who
were overweight or obese, but within white respondents, insurance coverage did not affect BMI.
Similarly, the proportion of persons reporting hypertension or diabetes differed across
race/ethnicity groups, but within a specific group, duration of insurance coverage had no
additional effects. The proportion of respondents reporting hypertension was highest among
12
African Americans (20.4%-24.4%); the proportion with a diabetes diagnosis was highest among
Hispanics (8.0%-8.8%).
Health Outcomes by Residence and Health Insurance Coverage
When characterizing respondents as “rural” or “urban,” we dichotomized the study
population, as noted earlier. Persons were defined as “rural” if they had lived ≥50% of the years
from 1989 through 1998 or 2000 outside metropolitan areas, and urban in all other cases (Table
B-5). While this definition is extremely broad, a tighter definition of rural yielded too few
persons for analysis of outcomes for minority populations. Health outcomes were tested between
people who were continuously covered and those who were not.
Within residence categories, but not controlling for other respondent characteristics,
continuous coverage was associated with significantly higher SF-12 Physical Component scores
for urban residents, but not for rural residents (See Figure 7 and Table B-5). The absence of
significance among rural respondents appears to stem from the larger differences associated
Figure 7. SF-12 Physical and Mental Component Scores at Age 40, by Residence and Health Insurance Coverage.
52.5
54.2
51.3 51.4
53.3 53.5
51.552.2
49505152535455
Physical SF-12 (Urban, p < .001) Mental SF-12 (both p < .001)
Rural, Continuous Rural, Not Continuous Urban, Continuous Urban, Not Continuous
with continuous insurance among urban persons. Note that the values for persons lacking
continuous coverage are the same among both rural and urban respondents (51.3 and 51.5,
13
respectively), while the improvement associated with continuous insurance is larger for urban
residents (53.3 for urban versus 52.5 for rural respondents). On the other hand, the proportion of
respondents reporting “fair to poor” health was nearly twice as high among persons who lacked
continuous health insurance coverage (15.7% rural, 15.4% urban) as among the fully insured
(8.6% rural, 8.3% urban). Findings from this measure were similar in direction and magnitude
across both rural and urban residents and were significant for both groups (Table B-5).
Examining SF-12 Mental Component scores within residence categories but not
controlling for other respondent characteristics, scores were significantly higher among
continuously insured persons in both urban and rural areas. Further, the effect size was larger
among rural residents (a 2.8 point difference) than among urban residents (1.3 point difference;
see Figure 7, previous page, and Table B-5).
The other health measures examined by the study, obesity and a reported diagnosis of
hypertension or diabetes, did not differ statistically by continuity of insurance coverage.
Holding Personal Characteristics Equal: Multivariable Analysis
The unadjusted analyses looked only at the effects of residence, race, and insurance.
However, other personal characteristics affect health, and may vary across the principal variables
of interest. The effects of residence, race and insurance, while holding other factors equal, were
studied using multivariable analytic models. In developing the modeling approach, we
attempted to include only variables that reflected events prior to, rather than simultaneous with,
the measurement of self-reported physical health at age 40. Thus, the insurance and residence
variables summarize status across the nine to eleven years preceding the interview (1989 through
1998 or 2000). Race, sex, education, marital status, and poverty during youth, are similarly,
long-term and/or prior statuses or events. An incremental modeling approach was used. The first
14
model examined the effects of the principal variables of interest--insurance, residence and race--
simultaneously on the relevant health outcome. The second model added sex, marital status,
education, and poverty status in 1979 to the preceding variables.
Factors Affecting Physical Health
Two general linear models were fitted for the SF-12 Physical Component scores (Table
B-6). Physical health scores for persons who were continuously covered were better than for
those who were not (52.8 vs 51.4, p=0.0001), adjusting for residence and race. Urban residents
had significantly higher scores than rural residents (51.5 versus 52.3, p = 0.0317), controlling for
insurance and race. Race was not a significant predictor of physical health scores when
residence and insurance were held equal.
After adjusting for sex, marital status, education, and poverty in youth, however, physical
health scores were no longer associated with continuous health insurance coverage or residence.
Education was highly associated with physical health status, with reported physical health
increasing with each increase in level of education. Males reported better physical health than
females. Married individuals had higher SF-12 physical health scores than did persons who had
never married. Persons who were poor in 1979 had slightly lower scores than those who were
not in poverty during youth (51.5 vs 52.3, p = 0.0369). Physician availability in the individual’s
present county of residence, when added to model 2, had no significant effects (results not
shown).
Factors Affecting Mental Health
The modeling approach used to study SF-12 Physical Component scores was also used to
analyze factors associated with SF-12 Mental Component scores (Table B-7). In the first model,
which adjusted for residence and race, individuals who had been continuously insured reported
better self-perceived mental health than others (mean 53.5 vs 52.7 and 51.8, p<0.0001). Mean
15
mental health scores did not differ with respect to residence or race when insurance status was
held constant.
Even when personal characteristics were held equal, individuals who had insurance
coverage for less than 75% of the time had worse mental health compared to the other two
groups, although the relationship was less strong (mean of 53.2 among continuously insured
versus 53.0 for 75%-99% and 52.3 among persons with ≤75%; p=0.0161). Differences for
residence or race were not significant. Males (54.3) reported better mental health than females
(51.6) and persons who were currently married reported better mental health than those who
were never married or in other statuses (mean of 53.3 versus 52.5 and 52.0, respectively; p =
0.0085). Education was to be associated with mental health, with persons having at least some
college education having better mental health status (p=0.0051). Poverty in youth tended to be
associated with lower SF-12 mental health scores (52.2 versus 53.1; p = 0.0520). In a third
model, physician availability was not significantly linked to mental health status at age 40
(p=0.6937).
Self Reported Health Status
Logistic regression was used to model whether a person would perceive their health
status as “fair” or “poor” (hereafter, poor health) versus better outcomes, using the same
incremental modeling approach as above (Table B-8). In a model assessing the simultaneous
effects of insurance, residence and race, the odds of poor health were higher for people covered
for less than 75% of the time and those covered 75%-99% (OR=1.85 and OR=1.42, respectively)
compared to persons who were continuously covered. Even with insurance coverage held
constant, the odds of reporting poor health were significantly higher for Hispanics (OR=1.62)
16
and African Americans (OR=1.61) compared to whites. Rural versus urban residence had no
effect.
With personal characteristics held constant, insurance coverage, residence and race were
not significantly associated with poor health at age 40. Strong education effects were present.
Individuals who had not completed 12th grade, or high school graduates with no further training,
were markedly more likely to report poor health than individuals who attended or graduated from
college (OR of 4.14 and 2.54, respectively, compared to college graduates; Table B-8). Males
were less likely to report fair or poor health (OR=0.75) than females. Marital status and poverty
during youth had no statistical effects. Physician availability, when added to model 2, had no
significant effects (results not shown).
Diagnoses and Health Risks
Obesity, a health risk variable that is associated with multiple co-morbidities, was studied
in a manner paralleling health status. When the effects of insurance, residence and race were
examined, rural residents (OR 1.36) and minorities (Hispanic, OR 1.51; African Americans, OR
1.98) were more likely to be obese at age 40 than urban residents or whites (Table B-9).
Continuity of insurance coverage was not associated with the risk of obesity.
When individual characteristics were added to the model, the relationship between rural
residence and obesity was no longer significant. Race remained significantly associated with
obesity, with Hispanics (OR 1.40) and African Americans (OR 1.84) having higher odds than
whites. Education was strongly linked to obesity, with persons with less than a college degree
being significantly more likely to be obese than graduates. While persons who never married did
not differ from married individuals with regard to risk for obesity, persons in the “other”
category, those who were widowed, separated or divorced, were significantly less likely to be
17
obese (OR 0.73). Sex and poverty during youth were not associated with obesity at age 40.
When added to the model, physician availability measured by physician rate per 100,000 persons
was not significant (p=0.4685; results not shown).
There was no association between diagnosis of hypertension at age 40 and insurance
coverage; this absence of effect was consistent across race and residence (Table B-10). Rural
residents were not more likely than their urban peers to report a diagnosis of hypertension.
African Americans were more likely to have a diagnosis of hypertension than were Whites
(OR=1.79); but Hispanics did not differ.
When controls for other personal characteristics were introduced in model two, results
were similar. Residence and continuity of insurance coverage were not associated with
diagnosed hypertension. The odds of hypertension remained higher among African Americans
than among whites (OR 1.58), while Hispanics did not differ from whites. No other factors were
associated with a diagnosis of hypertension by age 40.
Similarly, continuity of insurance coverage was not associated with a diagnosis of
diabetes at age 40; this effect was consistent across race and residence (Table B-11). Rural
residents were not more likely to report being diagnosed with diabetes. Hispanics (OR=2.31)
and African Americans (OR=1.78) were more likely to report a diagnosis of diabetes than were
whites. When personal characteristics were added to the model, insurance coverage and
residence still showed no relationship to diagnosed diabetes. The odds of diabetes remained
significantly higher among Hispanics (OR 2.22) than among whites, and approached significance
among African Americans. (OR 1.55, CI 1.00, 2.41) Males were less likely to report a
diagnosis of diabetes (OR 0.58) than were women. None of the other personal factors were
significantly related to a reported diagnosis of diabetes by age 40.
18
Chapter Three
Conclusions and Policy Recommendations Conclusions Effects of health insurance on physical health
In a population just reaching age 40, continuous health insurance coverage across the
preceding 9 to 11 years was not associated with better self-perceived health, as measured using
the SF-12 Physical Component Score. A positive relationship between insurance coverage and
health in unadjusted analysis, which remained after controlling for race and residence, ceased to
be significant after age, education, marital status, and poverty during youth were held constant.
Similar results were found when health status was dichotomized into “fair to poor” versus better
self-perceived health. Health insurance was not significantly related to the probability that a
respondent would report a high BMI (obesity), or having been diagnosed with hypertension or
diabetes. The relationship between insurance and physical health outcomes operated similarly in
rural and urban areas and across race/ethnicity categories.
Absence of the anticipated relationship between continuous insurance coverage and
physical health at age 40 may have multiple causes. First, respondents were aged 29 - 31 at the
beginning of the period studied. Previous longitudinal research into the effects of intermittent
insurance has focused on older populations, such as the persons aged 51 – 61 at baseline
followed by the Health and Retirement Survey (Sudano and Baker, 2003; Johnson and Crystal,
2000) or the Hispanic Established Populations for Epidemiologic Study of the Elderly (Angel,
Angel and Markides, 2002). Persons at younger ages may be less susceptible to disease in
general than older populations. Second, the study relied on patient report of diagnosed disease.
19
Persons who visit a health care provider regularly may be more likely to be diagnosed for
hypertension and diabetes, both of which are asymptomatic at early stages. Additional
limitations are detailed later in this chapter.
The principal drivers of perceived health status at age 40 were education (higher
education was associated with better health), marital status (married reporting better health), sex
(males reporting better health), and poverty during young adulthood (ages 19 – 21, associated
with poorer health). The latter effect is extremely interesting. Part of the effect of poverty while
young may stem from persistent poverty throughout adulthood; as noted, individuals who were
poor in 1979 were more likely to be poor in 1998-2000 than were others. Poverty rates were
reduced, but not eliminated, as persons first interviewed in 1979 matured. We did not include
poverty status at age 40 in the modeling approach, as we wished to avoid the endogeneity issues
associated with simultaneous measurement of the potential predictor and the outcome. Beyond
the continuation of poverty as a factor relating to poorer health, it is possible that health habits
and attitudes formed during youth persist despite subsequent financial improvement. An
emerging literature suggests that mortality is affected by childhood social status, in addition to
adult status (Poulton, Caspi et al 2002; Pensola and Martikainen, 2004; Beebe-Dimmer, Lynch et
al 2004). In an analysis of the National Longitudinal Survey of Older Men, financial status at
midlife was found to be more strongly associated with mortality than financial status in later life
(Hayward and Gorman, 2004). In addition, research is beginning to link childhood experiences
with adult morbidity (Blackwell, Hayward, Crimmins, 2001). Future research exploring duration
of time spent in poverty, paralleling the present research exploring time within insurance and
time spent in rural regions, is needed.
20
Effects of health insurance on mental health
While the proportion of time with insurance coverage was not significantly related to five
self-reported measures of physical health, continuous insurance coverage was associated with
better mental health, as measured by the SF-12 Mental Component Score. While the effect size
was small, in multivariable analysis it persisted even after residence, race, and demographic
covariates were held constant. Other factors associated with self-perceived mental health
paralleled those associated with physical health: sex, marital status, education and poverty
during youth. A direct causal relationship between insurance and mental health status cannot be
inferred, even though insurance is measured for an extensive period prior to the measurement of
mental health, however. The SF-12 Mental Component Score was not administered at the
initiation of the survey, in 1979, and thus it is possible that persons with better self-perceived
mental health at age 40 possessed that characteristic throughout their adult lives, and that better
mental health earlier in adulthood led to employment and other choices that resulted in
continuous insurance coverage from 1989 onward.
Effects of rural residence
Among persons with similar coverage, the relationship between health insurance
coverage and physical and mental health outcomes was not affected by residence. Among both
urban and rural residents, persons lacking continuous insurance reported lower SF-12 physical
health scores, although the relationship was not statistically significant among rural respondents.
The proportion of persons reporting fair to poor health status was statically higher for persons
without continuous health insurance coverage among both urban and rural residents. In
21
multivariate analysis, however, the effects of rural residence on self reported physical and mental
health were not significant after other personal characteristics of respondents were held equal.
While the effects of insurance did not differ between rural and urban respondents, it is
worth noting that many risk factors associated with poorer health status, such as low education
levels and poverty, have been found to be more common in rural populations (Glover, Moore et
al, 2004).
Effects of Minority Status
With differences in insurance coverage, residence, education, marital status, age, and
poverty in young adulthood held constant, neither African Americans nor Hispanics in the study
perceived themselves to have poorer physical or mental health than whites. The only significant
differences that emerged pertained to obesity, a risk factor, and hypertension and diabetes,
diagnoses associated with obesity. As expected, obesity and diabetes were more common
among both African Americans and Hispanics than among whites (Tables 4, 8 and 10). For
diabetes, race/ethnicity was the only predictive factor that was significant in multivariable
analysis (Table 10). Hypertension was more likely among African Americans than whites, but
odds did not differ for Hispanics. Findings regarding increased prevalence of diabetes and
hypertension among minorities reflect national trends for these populations (Lucas, Schiller and
Benson, 2004; Mensah, Mokdad, Ford, Greenlund and Croft, 2005).
It is interesting that although the presence of diagnosed disease was significantly higher
among minority respondents, they did not report lower levels of physical health, other factors
held equal. These results may be due in part to the relatively young age of respondents. The
oldest individuals in our sample were 40. Thus, more serious health complications have likely
not yet emerged, even among persons with diagnosed disease.
22
Effects of Education
In light of continuing debates on the adequacy of state funding formulas that allocate
resources to rural schools, it is worthwhile examining the effects of education on health at age
40. Educational attainment, measured in four increments from less than high school through
college graduation or above, was not significantly related to the two specific diagnoses studied,
diabetes and hypertension, in adjusted analyses. However, it was strongly related to the risk of
being obese. Further, self-perceived physical and mental health scores increased steadily with
education. The effect on self-reported ‘poor to fair’ health status was striking. The odds that an
individual who did not graduate from high school would report poor health at age 40 were 5.42
times higher than those of college graduates. Increased education may be associated with
occupations less likely to cause physical stress and strain, with increased self-care skills, or
simply with different definitions of expected health.
Health Insurance Coverage
While examining the effects of health insurance was the principal purpose of the study,
levels of insurance observed in the population are also worth discussing. Over the 1989 –
1998/2000 period, only slightly more than a third of rural minority respondents (37%) were
continuously insured. While whites were more likely to be continuously insured (57%) than
minorities, the proportion is both objectively low and statistically lower than among urban
whites.
The absence of health insurance does not appear to have had adverse effects on the
respondents’ self-perceived physical health at age 40. However, low rates of coverage are likely
to continue as the cohort ages. A recent analysis of insurance coverage for the period 1992 –
2000 among older adults (51 – 57 at baseline) found that 21.8% were uninsured at one or more
23
of five survey administrations across the period (Baker, Sudano, 2005). Further, minorities were
at higher risk than whites for both a single uninsured period and, particularly, for being
uninsured at two or more survey administrations. Rising premiums are projected to increase over
the next 10 years, potentially further exacerbating gaps in coverage (Gilmer, Kronick, 2005).
Thus, as the NLSY study population ages, disparities between insured and uninsured
populations, and across racial/ethnic groups, may be anticipated to worsen.
Limitations to the Present Analysis
The analysis performed for this report is innovative. To the best of our knowledge, no
previous researchers have explored materials from the National Longitudinal Survey of Youth
initiated in 1979 to ascertain effects on health of exposures such as insurance coverage and rural
residence. However, the initial research on a topic makes assumptions and creates definitions
that may be improved by subsequent work. Limitations to the present analysis include:
• Defining rural residence. Several limitations surround the way “rural” was defined.
Although the degree to which the US population as a whole is “mobile” has been
questioned (Wolf and Longino, 2005), very few persons remained in non-
metropolitan settings for the approximately 9-11 years they were studied. Therefore,
we chose to express residence as a proportion of time, and to dichotomize at the 50%
point. Both the use of percentages and the specific cut-points used can be debated.
A definition of “rural” that was restricted to persons living exclusively in rural areas
would be theoretically more powerful. However, in the dataset used, the small
“completely rural” population would have meant less power for the analysis. An
additional limitation is caused by the absence of further delimitations within “rural.”
The health status of an individual who spends early adult life in a frontier area may
24
differ from that of a person living in a large rural county with significant urbanized
populations.
• Defining insured. Parallel limitations are present in the definition of coverage. While
the analysis specifies percent of time insured between 1989 and 1998/2000, it does
not distinguish between persons whose period of uninsurance was early in the study
period and those for whom it might have been recent, or between a single extended
period of uninsurance and multiple such periods. However, the definition of
“continuous coverage” is fairly straightforward, and the time span covered is greater
than that used by previous research. Second, the report does not distinguish between
private and public insurance coverage. Several previous studies have shown that
adults with public health insurance generally have poorer health status than the
privately insured, and may have transferred from private insurance of an uninsured
state in response to such poor health (Baker and Sudano, 2005). Grouping these two
categories may have obscured differences in health between the privately insured,
assumed to have better health than the composite group, and those lacking insurance
coverage. The modeling approach may also be a limitation. “Continuous coverage”
was used as the referent group. Different effects may have been reached had we used
the small group of persons who never had insurance at any time. Later analyses, as
more of the NLSY-79 cohort reach age 40, may be able to employ this approach.
• Age of respondents. As noted earlier, age 40 is relatively young, as far as the health
effects of chronic disease are concerned.
• Outcomes. All health outcomes used in the study are based on respondent self –
report, even though two are drawn from recognized instruments. Absent clinical
25
examination, an objective assessment of health status cannot be obtained. The most
likely result is that the true prevalence of disease in the population is under-estimated.
Policy Recommendations
Although health insurance, in the present analysis, was not related to most measures of
physical health status once demographic and socioeconomic variables were taken into account,
we are reluctant to conclude that health insurance is not necessary, or not needed until
individuals reach older ages. It seems likely that the individuals examined in this study were too
young for major adverse health events and major declines in health to occur. Nearly half (46%)
of the population studied here did not have continuous insurance coverage between the ages of
29-31 and age 40. If NLSY respondents reacted to insurance gaps as did individuals in previous
research, they are likely to have deferred routine care during their period without insurance
(Sudano, Baker 2003), or to have problems meeting medical expenses (Schoen, DesRoches
2000) and to have sought public insurance or charity care when hospital care was needed
(Johnson, Crystal 2000). Further, the population studied will continue to be at risk for gaps in
health insurance coverage as they age (Baker and Sudano, 2005)
Mental health was the only condition that was significantly related to continuous health
insurance coverage. This is an important finding: mental health has the second largest burden of
disease in most market economies including the U.S., second only to cardiovascular diseases
(NIMH, 2001). Although there has been little research on the costs of mental illness, it is
assumed that the direct and indirect costs associated with mental illness, treatment, and recovery
are severe. Missed work, inability to carry out daily tasks, long treatment periods and expensive
care all contribute to the cost of mental illness. The Mental Health Parity Act of 1996 (MHPA)
may have contributed to the association between continuous insurance coverage and improved
26
mental health. Since 1998, companies with 50 or more employees have had to offer comparable
benefits for mental health and physical health if they offered coverage for mental health services
(CMS 2005). Despite the many loopholes written into the MHPA, its extension from the year
2000 into 2004 indicates that lawmakers and administrators perceive that the law has had some
success in improving mental health coverage. Given the link between mental health status and
continuous insurance coverage found in this study, future research should try to quantify the
impact of the MHPA on the extent of coverage for mental health services and use of mental
health services.
Although there were no significant differences between rural and urban adults in the
effect of continuous insurance on health status, it is interesting to note that rural residents overall
were less likely to be continuously insured than urban residents. In addition, rural minorities
were particularly disadvantaged with respect to having continuous insurance coverage. Given
our continuing belief that insurance will play a significant role in improving health status as
these individuals age, programs that target improved insurance coverage and access to care
among rural residents, especially minorities, will continue to be needed. In Andersen’s
Behavioral Model, health insurance is an enabling factor that can improve health status by
reducing financial barriers to accessing care. However, as discussed below, the difficulty of
expanding insurance coverage makes other methods of improving access equally important.
Expanding Medicaid coverage is not a viable option at this time. First, the FY2006
Federal Budget recently passed by Congress calls for a reduction in the Federal contribution to
Medicaid (Abrams, 2005). Under the budget resolution, $10 billion would be cut from Medicaid
over the 2007 – 2010 fiscal years. Even before the new Federal budget stance, many states were
facing budget crises that threaten existing coverage and have implemented or plan to implement
27
cost containment policies (Smith, Ramesh, Gifford, Ellis, Rudowitz & O’Malley, 2004). From
2002 to 2004, states exercised a variety of cost-control options, from reducing eligibility and
benefits and increasing co-payments on the demand side, to reducing or freezing provider
payments and controlling drug costs on the supply side. These same strategies are being
considered by other states for FY 2005. Enrollment in Medicaid has continued to grow based on
existing eligibility standards, while state tax revenue has grown at a much slower rate than
growth in Medicaid costs (Smith, Ramesh, et al, 2004).
Mental health care presents additional problems, particularly as regards a rural mental
health safety net (New Freedom Commission, 2004). For a variety of reasons, including
managed care, downsizing, consolidation, and state budgetary constraints, inpatient
hospitalization and residential treatment decreased demonstrably during the 1990s. This
paradigm shift has resulted in persons with persistent and severe mental illness being treated in
community-based settings, such as Community Mental Health Centers. With Community
Mental Health Centers caring for more persons with high levels of acuity, individuals with less
severe conditions, such as persons in the current study, are not treated due to the capacity issues.
Overall, programs that target expanded access to the uninsured without changing their
insurance status may offer the most feasible solutions to the provision of physical and mental
health care. The United States’ health policy works through incremental reform (Short, Shea,
Powell, 2003); therefore, it would be most productive to encourage planners to think of short-
term approaches that can fill gaps, as well as more over-arching solutions (Short & Graeffe,
2003). However, it should be noted that communities most in need of services often lack strong
safety net institutions, and thus are unable to compete successfully for grant funding (Hoadley,
Felland and Staiti, 2004).
28
Recommendation: The Department of Health and Human Services should evaluate how
its component agencies can best support local programs working to improve access to
care for uninsured populations. Working with professional organizations, including the
American College of Healthcare Executives, the American Hospital Association, and the
National Rural Health Association, DHHS should ensure that best practices among such
local efforts are identified and shared. Technical assistance materials developed to
document best practices should be available to, and understandable by, non-specialist
community leaders.
Federally Qualified Community Health Centers (CHCs) and Rural Health Clinics
(RHCs): CHC’s and RHCs may offer an option for expanding access to health care to rural
residents without changing their insurance status. CHCs have grant support that allows them to
charge sliding scale fees to uninsured persons. However, awareness of this program among the
general public may be limited. At present, RHCs receive expanded reimbursement for Medicare
and Medicaid patients, but receive no specific support to help them address the needs of un- and
under-insured patients. Local collaborations and referral arrangements between CHCs and
RHCs offer one potential means of directing uninsured persons to a low-cost source of care.
Recommendation: The Secretary of the Department of Health and Human Services
should work with the National Association of Rural Health Clinics, the National Rural Health
Association, and the National Association of Community Health Centers to define model
approaches to involving RHCs in the care of uninsured persons. It would be particularly
valuable to document collaborations among RHCs, CHCs, and other service providers, and to
communicate the information broadly to state and local planners.
29
Rural Practitioner Skills. The decreased mental health reported by respondents who
lacked continuous health insurance was not sufficiently severe as to result, for example, in
prolonged hospitalization that might have led them to drop out of the survey. Nonetheless,
recognition of problems by health care providers might have alleviated distress. Because
expansion of mental health safety net services through additional mental health practitioners is
not feasible in all areas, ensuring that primary care providers are knowledgeable offers a short-
term solution. Training resources could be offered to state medical societies for use in
continuing medical education programs.
Recommendation: In partnership with local mental health professional organizations,
Area Health Education Consortia are encouraged to develop strong educational programs
in mental health screening, treatment, and referral for family medicine, internal medicine,
and general surgery resident physicians, and for nurses. Such educational content is
needed both during practitioner training and as part of continuing medical education
programs.
Improving practitioner pipelines. Educational attainment was strongly related to all health
outcomes. The health insurance status of individuals possibly improved, and the pipeline of
future practitioners enhanced, through efforts to recruit a broad range of individuals into health
careers. Re-education, such as that provided through the Workforce Investment Act of 1998
(WIA), can serve as a mechanism for addressing current low income, uninsured persons by
allowing them to enter careers with better prospects for pay and benefits. A recent study of
programs funded by the WIA found that multiple states were using the program to develop
health careers, both among youth and among low-income adults (Skillman et al, 2005)).
30
Recommendations:
• State Area Health Education Consortiums are encouraged to partner with local
middle and high schools in an attempt to promote health careers, both as a means
of encouraging children to higher educational attainment and as a way of growing
a local health labor force. Such activities should be particularly promoted in rural
areas.
• State Area Health Education Consortiums are encouraged to partner with agencies
in their state responsible for implementation of the WIA, in order to identify
health careers job training opportunities that may benefit rural residents and
institutions.
Research Recommendations
Research recommendations flow directly from the findings, and the limitations, of the
present study:
- The youngest members of the NLSY-79 cohort reach age 40 in 2005. When
public use data are available for the full NLSY-79 cohort, the analysis reported here should be
repeated. With larger sample sizes, many limitations of the present analysis may be obviated.
For example, more detailed definitions of “rural” and “insured” will be possible.
- Poverty at ages 19 – 21 appears to have lasting effects on perceived mental and
physical health, even controlling for education and health insurance. This long-term effect may
result from culture as learned in youth, from employment choices, or from other factors not
included in the present analysis. Identification of factors which ameliorate the effect of early
31
poverty, which may include education, employment in varying occupations, and so on, is
essential.
32
Appendix A: Methods
The National Longitudinal Survey of Youth The National Longitudinal Survey of Youth (NLSY-79), commissioned by the Bureau of
Labor Statistics, is one of the nation’s longest running panel surveys (NLSY79, 2001). A
multistage, stratified clustered sample, it was initiated in 1979 with three components: a sample
representative of the population of persons born between 1957 and 1964; supplemental samples
of African Americans, Hispanics, and economically disadvantaged whites; and a sample of
active duty military. The latter two groups, military and disadvantaged whites, were dropped
from follow-up after 1990.
Participants were interviewed annually from 1979 through 1994, and biannually until
present, with an 84% retention rate through 1998. Retention rates do not differ by race. In all
except one survey round (1987), most interviews were conducted in person, with some
individuals who were difficult to reach interviewed by telephone. A Spanish version of the
questionnaire was developed and was used if the respondent requested it.
While principal focus of the NLSY has been labor and economic outcomes, information
about life experiences (marriage, residence) and health were also obtained. An additional survey
(NLSY79 Children) has tracked growth and development among children of the originally
sampled individuals.
Throughout the years, information on the health of each participant has been collected to
ascertain the individual’s fitness for labor force participation. This information has included
height, weight and health conditions that would prevent the person from being active in the labor
market. Beginning in 1998, as the first group of the cohort reached the age of 40, an additional
section was added to the interview that addressed respondent health. This section included the
mental health and physical health components of the SF-12 along with self-reported health status
and other health outcomes measures.
The analysis presented in this report is limited to the subset of NLSY-79 participants
(3,137 persons) who had reached the age of 40 by 2000 and had less than 3 missing interviews
between 1989 and 2000, the most recent administration of the NLSY-79 available in a public-use
data set. For this subset of individuals, we examined information from all administrations of the
33
NLSY from the initial survey through the survey year in which respondents turned 40 and
received the health interview. The outcomes studied were the measures of health status used by
the NLSY: self-reported health status, SF-12 physical and mental component scores, body mass
index (BMI), and a reported diagnosis of hypertension or diabetes.
Insurance Status
From 1979 on, respondents were asked if they received health insurance as an employee
benefit. However, a “no” answer to this question does not imply that the respondent was
uninsured, as insurance may be privately purchased or the respondent could be covered under a
spouse’s policy. Further, the question was not asked of respondents who were not in the work
force at the time of the interview. Thus, women who were temporarily out of the workforce to
care for children would have been excluded. In 1989 and subsequently, the insurance questions
were expanded to ask the respondent explicitly about his or her coverage. Respondents were
asked annually (biannually after 1994) whether they were insured at the time of the interview.
Thus the NLSY-79 data set, with 7-8 observation points across 12 years, provides substantial
historical information about insurance coverage, allowing us to examine current health
conditions in light of previous, rather than simultaneous, insurance coverage. The specific
NLSY-79 questions used for the present study to define coverage are in the table below.
Table A-1. Source questions for insurance variable Year Question Response 1989 Who is covered by a health/hospitalization plan
(respondent) Respondent
1990 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
1992 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
1993 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
1994 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
1996 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
34
1998 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
2000 Are you covered by any kind of private or government health or hospitalization plans or health maintenance organization (HMO) plans?
Yes/No
We created a measure indicating the percentage of coverage an individual had across the
10-12 year period by summing up the number of years with a ‘Yes’ response to the above
questions and then dividing by the number of years surveyed. Participants were not included in
the analyses if a respondent had missing responses for 3 or more years. For those persons with 1
or 2 missing responses, imputation was used for the missing values using a random binomial
number generator in SAS (Cary, NC).
Residence
In a mobile society, determining the effects of rural residence over a 20-year period is not
straightforward. Across the study population, from 34% (whites) through 16% (Hispanics) of
respondents were living in rural areas, defined as outside standard metropolitan areas, in 1979.
By 2000, the last survey year, these values had dropped to 10% for whites and less than 4% for
Hispanics. Further, a substantial portion of NLSY participants moved during the study period.
Because we were examining the effects of insurance over time, we similarly defined
“rural residence” as a function of time. For each individual, we calculated the percentage of time
spent outside a standard metropolitan area. For dichotomous analyses, we defined “rural
resident” as a person who had lived less than half of the studied period in an urban area.
Health Status
Several health outcomes were studied. The first set of measures was scores from the
Short Form 12 (SF-12) survey, Physical Component Scores (PCS) and Mental Component
Scores (MCS). These scores can range from 0-100 and are designed to have a mean value of 50
in a representative US population (Ware, 1996).
Self-reported health status was also used as an outcome, based on a question within the
SF-12. Respondents could describe their own health as Excellent, Very Good, Good, Fair or
35
Poor. Responses were grouped into two categories, Excellent, Very Good or Good versus Fair or
Poor.
Body mass index (BMI) was considered a health outcome and was analyzed as both a
continuous variable and as a dichotomous variable. Obesity was defined as having BMI greater
than 30.
The NLSY-79 health survey asks respondents whether they have experienced an
extensive list of conditions, from cancer and congestive heart failure through allergies. In
determining which of these conditions to use, we focused on conditions generally believed to be
preventable through primary care, excluding diagnoses such as cancer for which the etiology is
less well known. In addition, we limited the analysis to conditions with sufficient prevalence in
the study population for adequate analysis. This removed from consideration conditions such as
stroke and congestive heart failure, which have very low frequency in a population aged 40. The
specific conditions selected for analysis were hypertension and diabetes, which could in many
cases be avoided through preventive counseling regarding weight control, diet, and exercise, but
which occurred frequently enough for analysis. Smoking status was not used as an outcome for
the study because the questions from 1998 and 2000 were not similar and could not be
combined. In 1998, a question was asked pertaining to smoking daily, occasionally, or not at all.
In 2000, smoking questions were only addressed to women who had been pregnant.
Other Determinants of Health Outcomes
In addition to insurance coverage and area of residence, resources in the community in
which a respondent was currently living could affect his or her health outcomes. Persons with
poor availability of health care resources, or living in a generally poor community, may not
receive health care that equals that of residents in richer communities. To capture the effects of
current resources on health, we included a measure for number of physicians per 100,000
persons (1998 or 2000). We also examine unemployment rate in the county of residence, but it
proved to have no relationship to any outcomes.
36
Statistical Methods
In original 1979 sample, Hispanic and Blacks were over-sampled therefore weights
would need to be applied for analyses that summed across race/ethnicity. For descriptive
percentages, we stratified by race and rural, providing no estimates for the total population.
Therefore, no sampling weights were needed.
Sampling weights from a specific year (1998 or 2000) were applied only when the
analyses were not stratified by race. For all modeling, we controlled for race/ethnicity (sampling
strata) and took into account design effects from the original sample design in 1979.
Chi-square analyses were used to compare percentages of demographic variables across
races, to compare percentages of race and residence across coverage, and to compare percentages
of categorical health outcomes by race, residence, and coverage. Means and standard errors
were presented for SF-12 scale components, physical health and mental health. Means were
tested for differences using analysis of variance (ANOVA).
Logistic regressions were used to determine the odds of having defined categorical health
outcomes (Fair to Poor health status, obesity, hypertension, and diabetes). Because of the
continuous nature of both the SF-12 physical component and mental component scores, these
scores, used as dependent variables, were analyzed again using general linear modeling. This
allowed estimating the least squares means for each level of the independent variables.
All modeling started with a model where the outcome was predicted by health insurance
coverage as a 3 level variable (100% coverage, 75-99% coverage, >75%), race (White, African
American, Hispanic), residence (>50% in urban, >50% in rural), and two interaction terms, one
between insurance and race and the other for insurance and residence. These allowed us to test if
the effect of insurance coverage on the health outcomes differed by residence or by race. Once
the significance of the interactions was determined, the next model included other control
variables, sex, marital status, education, and poverty status in 1979. A third model was run to
assess the effects of physician availability, measured at the time the health survey was conducted
(1998 or 2000). Because this variable was generally not significant, and did not substantially
change values for other variables in the model, results are not shown.
37
Appendix B: Tables Table B-1. Description of Population, Health Interview at Age 40 N=3,104 Hispanic
n=571 Black n=964
White n=1569
Chi-square p-value
% (SE) % (SE) % (SE) Female 51.8 (1.9) 53.8(1.9) 53.3 (1.3) 0.7270 Education 0.0000
Less than 12th Grade 22.8(1.8) 15.2 (1.5) 7.7 (0.8) High School Graduate 37.3 (2.4) 43.9 (1.8) 43.1 (1.6) Some College 26.4 (1.9) 27.1(1.7) 23.7 (1.2) College Graduate or Greater
13.5 (1.5) 13.7 (1.3) 25.5 (1.6)
Marital Status 0.0000 Never Married 14.7 (1.7) 32.9 (1.7) 9.4(0.8) Currently Married 60.9 (2.0) 36.5 (1.8) 72.5 (1.0) Other 24.3 (1.8) 30.5(1.5) 18.0 (0.9)
In Poverty in 1979 28.8 (2.2) 38.9 (2.3) 9.9 (0.8) 0.0000 In Poverty at age 40 15.2 (1.8) 24.8 (1.5) 7.5 (1.0) 0.0000 Residence: Percent Time in SMSA
0.0002
0-25% 6.7 (2.0) 12.1 (3.7) 20.3 (2.9) >25% - 50% 1.4 (0.5) 1.9 (0.4) 4.5 (0.6) >50% - 75% 6.8 (2.5) 5.2 (1.8) 6.4 (1.2) > 75% 85.1 (3.3) 80.8 (4.0) 68.8 (3.1)
Point residence estimates: Not in SMSA 1979 16.1 (4.7) 25.9 (5.4) 34.6 (4.5) 0.0158 Currently Not in SMSA (1998 or 2000)
3.5 (0.9) 5.2 (1.3) 9.8 (1.1) 0.0001
County of residence, 1998 or 2000
MD population rate (per 100,000)
240.1 (12.1) 307.2 (15.4) 226.2(8.3) 0.0000
Unemployment rate in county of residence
6.3 (0.5) 4.5 (0.1) 4.3 (0.2) 0.0003
38
Table B-2. Distribution of health insurance coverage by race, residence (1979 and current), and residence over time. Continuously
covered Coverage 75 – 99%
Coverage 50 – 74%
Coverage 25-49%
Coverage 0 – 24%
%(SE) %(SE) %(SE) %(SE) %(SE) Total 53.6 (1.3) 14.8 (0.7) 14.9 (0.8) 9.7 (0.6) 7.0 (0.6) Race* White Black Hispanic
63.9 (1.5) 43.2 (1.8) 42.9 (2.0)
12.2(0.9) 18.3(1.4) 15.9(1.5)
10.3 (0.9) 19.1 (1.3) 19.8 (1.9)
8.1 (0.8) 10.9 (1.0) 11.9 (1.4)
5.5 (0.7) 8.2 (0.8) 9.5 (1.5)
Residence baselinea
Rural Urban
57.7 (2.3) 61.1 (1.7)
13.8 (1.4) 12.8 (0.8)
11.1 (1.2) 12.1 (0.9)
8.7 (1.4) 8.8 (0.8)
8.7 (1.5) 5.2 (0.6)
Current residencea
Rural Urban
56.4 (3.3) 60.1 (1.5)
15.6 (2.6) 13.1 (0.7)
12.4 (2.3) 12.4 (0.8)
8.5 (2.1) 8.8 (0.7)
7.1 (2.1) 6.0 (0.6)
Residence (% urban) 0 - 25% 26-50% 51-75% 76-100%
56.1 (3.3) 50.9 (5.9) 45.4 (4.3) 62.3 (1.5)
15.2 (1.8) 9.5 (3.5) 10.5 (2.3) 13.3 (0.8)
11.3 (1.6) 17.4 (4.1) 19.3 (3.2) 11.4 (0.9)
9.3 (1.7) 11.9 (3.4) 13.5 (3.6) 8.0 (0.7)
7.9 (1.8) 10.2 (3.4) 11.3 (2.4) 5.1 (0.5)
a For people turning 40 in 1998, sampling weight is from 1998; for people turning 40 in 2000, sampling weight is from 2000. P values for non-significant comparisons:
Residence baseline p=0.2652, Current residence p=0.7813.
* p <0.0001 ** p=0.0012
39
40
Table B-3. Outcomes by health insurance coverage*, ** Continuously
covered (54.0%; n=1694)
Coverage 75 – 99% (14.6%; n=458)
Coverage 50 – 74% (14.7%; n=462)
Coverage 25-49% (9.7%; n=303)
Coverage 0 – 24%
(7.05; n=220)
Mean (SE) Mean (SE) Mean (SE) Mean (SE) Mean (SE) SF-12 Physical Component
53.1 (0.2)
51.7 (0.4)
51.5 (0.5)
51.2 (0.6)
51.1 (0.7)
SF-12 Mental Component
53.6 (0.2)
52.6 (0.4)
51.0 (0.6)
51.8 (0.6)
52.9 (0.6)
% (SE) % (SE) % (SE) % (SE) % (SE) Fair to Poor Health
8.3 (0.8)
13.5 (1.9)
16.1 (2.0)
16.7 (2.4)
17.0 (3.6)
BMI % Overweight % Obese
36.7 (1.4) 25.9 (1.3)
37.2 (2.9) 28.2 (2.4)
36.7 (3.3) 28.6 (3.1)
29.2 (3.1) 30.7 (3.1)
36.1 (4.0) 22.9 (3.6)
HTN Diagnosis 13.7 (0.9) 16.8 (1.9) 20.4 (2.4) 12.0 (2.1) 16.1 (3.3) Diabetes Diagnosis
4.0 (0.5)
4.1 (1.1)
6.6 (1.3)
3.2 (1.1)
2.2 (1.1)
*For people turning 40 in 1998, sampling weight is from 1998; for people turning 40 in 2000, sampling weight is from 2000. **P-values, measuring overall differences across the five coverage categories:
SF-12 Physical Component p=0.0002; SF-12 Mental Component p=0.0000; Fair to Poor Health p=0.0001; BMI p=0.3007; Hypertension p=0.0855; Diabetes p=0.1395.
Table B-4. Health Insurance and Health Outcomes Stratified by Race N=3,137
Hispanic n=575
Black n=963
White n=1,599
Not Continuous Continuous Coverage
Not Continuous Continuous Coverage
Not Continuous Continuous Coverage
Mean (SE)
Mean (SE)
Mean (SE)
Mean (SE)
Mean (SE)
Mean (SE)
SF-12 Physical Component
50.9 (0.5) A
52.7 (0.5)
51.3 (0.4)
52.1 (0.4)
51.4 (0.4) A
53.2 (0.3)
SF-12 Mental Component 52.7 (0.5) 53.6 (0.5) 51.9 (0.4)B 53.0 (0.4) 51.9 (0.4)B 53.7 (0.2) BMI Mean (SE) 28.5 (0.3) 28.3 (0.4) 28.9 (0.3) 29.7 (0.3) 27.0 (0.3) 26.9 (0.2)
% (SE)
% (SE)
% (SE)
% (SE)
% (SE)
% (SE)
Self-reported Health Status: Fair to Poor
19.1 (1.9)C
12.2 (1.7)
18.3 (1.8)C
13.5 (1.7)
14.3 (1.5)C
7.4 (0.9)
BMI % Overweight % Obese (BMI ≥30)
41.1 (2.6) 32.7 (2.6)
39.9 (2.8) 30.3 (2.9)
37.5 (2.2) 36.5 (2.0)
36.2 (2.5) 40.7 (2.6)
32.8 (2.4) 25.4 (1.9)
35.7 (1.6) 24.3 (1.4)
HTN Ever Diagnosed 14.8 (1.9) 13.5 (2.2) 24.5 (1.5) 20.5 (1.7) 14.9 (1.7) 13.1 (1.1) Diabetes Ever Diagnosed 8.0 (1.5) 8.2 (1.8) 6.6 (1.2) 6.0 (1.1) 3.5 (0.7) 3.7 (0.6) A Hispanic Not Continuous vs Continuous p=0.0135; White Not Continuous vs Continuous p=0.0001 B Black Not Continuous vs Continuous p=0.0349; White Not Continuous vs Continuous p=0.0000 C Hispanic Not Continuous vs Continuous p=0.0073; Black Not Continuous vs Continuous p=0.0612; White Not Continuous vs Continuous p=0.0001
41
42
Table B-5. Health Insurance and Health Outcomes Stratified by Residence
Rural ≥ 50% n=607
Urban ≥ 50% n=2530
Not Continuous
Continuous Coverage
Not Continuous
Continuous Coverage
Mean (SE) Mean (SE) p-value Mean (SE) Mean (SE) p-value SF-12 Physical Component
51.3 (0.6) 52.5 (0.5) 0.1701 51.5 (0.3) 53.3 (0.3) 0.0000
SF-12 Mental Component
51.4 (0.6) 54.2 (0.4) 0.0000 52.2 (0.3) 53.5 (0.2) 0.0005
BMI Mean (SE) 27.7 (0.5) 28.1 (0.4) 0.5009 27.5 (0.2) 27.0 (0.2) 0.0593 % (SE) % (SE) % (SE) % (SE) % (SE) % (SE) Self-reported Health Status: Fair to Poor
15.7 (2.7)
8.6 (1.9)
0.0176
15.4 (1.4)
8.3 (0.9)
0.0000
BMI % Overweight % Obese (BMI ≥30)
33.1 (4.0) 28.7 (3.7)
37.7 (3.4) 32.2 (2.9)
0.2440
35.8 (1.8) 27.8 (1.5)
36.5 (1.6) 24.5 (1.4)
0.1703
Hypertension Ever Diagnosed
15.2 (2.1) 15.2 (2.0) 0.9702 17.2 (1.5) 13.3 (1.1) 0.0544
Diabetes Ever Diagnosed 2.6 (1.2) 4.8 (1.2) 0.2052 4.9 (0.7) 3.8 (0.6) 0.2324
Table B-6. General Linear Model Results for SF-12 Physical Component Scores. Model 1* Model 2** LS Mean (SE) p-value LS Mean
(SE) p-value
Insurance Coverage
≤ 75%
51.4 (0.3)
0.0001
51.9 (0.3)
0.1848
75%-99% 51.4 (0.4) 51.6 (0.4) 100% coverage 52.8 (0.2) 52.4 (0.2) Residence >50% in Rural 51.5 (0.4) 0.0317 51.7 (0.3) 0.2069 >50 % in Urban 52.3 (0.2) 52.2 (0.2) Race Hispanic 51.8 (0.4) 0.0696 52.0 (0.4) 0.8188 Black 51.8 (0.3) 52.3 (0.3) White 52.5 (0.2) 52.1 (0.2) Sex Male - - 52.9 (0.2) 0.0000 Female - - 51.4 (0.2) Marital Status Never Married - - 51.3 (0.4) 0.0024 Married - - 52.6 (0.2) Other - - 51.5 (0.4) Education <12th Grade - - 49.5 (0.6) 0.0000 High School Grad. - - 51.9 (0.2) Some College - - 52.8 (0.3) College Grad. + - - 53.5 (0.3) Poverty 1979 Not in Poverty - - 52.3 (0.2) 0.0369 Poverty - - 51.5 (0.3) *Model 1 originally included two interaction terms (insurance and race, insurance and SMSA status). The effect of insurance did not differ by race (insurance * race, p=0.4052) or SMSA status (insurance * SMSA, p=0.6021). **Model 3 was run with all variables in Model 2 with the addition of unemployment rate and physician rate per 100,000 population. None of the least square means for the variables included in Model 2 changed by more than 0.1. Neither unemployment in the respondent’s county of residence at the time of the survey (p = 0.2247) or physician rate was a significant predictor of physical health (p=0.0745).
43
Table B-7. General Linear Model Results for SF-12 Mental Component Score Model 1* Model 2** LS Mean
(SE) p-value LS Mean (SE) p-value
Insurance Coverage
≤ 75% 51.8 (0.3)
0.0000
52.2 (0.3)
0.0161
75%-99% 52.7 (0.4) 53.0 (0.4) 100% coverage 53.5 (0.2) 53.3 (0.2) Residence >50% in Rural 53.0 (0.3) 0.6387 53.0 (0.4) 0.6317 >50 % in Urban 52.8 (0.2) 52.8 (0.2) Race Hispanic 53.2 (0.4) 0.3713 53.4 (0.4) 0.2592 Black 52.6 (0.3) 53.0 (0.3) White 52.9 (0.2) 52.8 (0.2) Sex Male - - 54.3 (0.2) 0.0000 Female - - 51.6 (0.3) Marital status Never Married - - 52.5 (0.4) 0.0085 Married - - 53.3 (0.2) Other - - 52.0 (0.4) Education <12th Grade - - 51.4 (0.6) 0.0051 High School Grad. - -
52.7 (0.3)
Some College - - 53.6 (0.3) College Grad. + - - 53.2 (0.3) Poverty 1979 Not in Poverty - - 53.1 (0.2) 0.0520 Poverty - - 52.2 (0.4) *Model 1 originally included two interaction terms (insurance and race, insurance and SMSA status). The effect of insurance did not differ by race (p=0.7860) or SMSA status (p=0.3431). **A third model was run including all variables in model 2 and the addition of unemployment rate and physician rate per 100,000 population. None of the least square means for the variables included in Model 2 changed by more than 0.1. Neither unemployment in the respondent’s county of residence at the time of the survey (p = 0.2772) or physician rate was a significant predictor of physical health (p=0.6937).
44
Table B-8. Logistic Regression Results for Factors Associated with Self-reported Fair or Poor Health Status Model 1* Model 2** O.R. 95% C.I. O.R. 95% CI Insurance Coverage ≤ 75% 1.85 1.44, 2.33 1.29 0.97, 1.73 75%-99% 1.42 1.08, 1.99 1.14 0.83, 1.58 100% coverage 1.00 ----- 1.00 ----- Residence >50% in Rural 1.20 0.88, 1.63 0.99 0.73, 1.34 >50 % in Urban 1.00 ----- 1.00 ----- Race Hispanic 1.62 1.22, 2.15 1.28 0.94, 1.74 Black 1.61 1.24, 2.10 1.22 0.91, 1.63 White 1.00 ---- 1.00 ---- Sex Male - - 0.59 0.46, 0.75 Female - - 1.00 ---- Marital status Never Married - - 1.36 1.00, 1.55 Married - - 1.00 ---- Other - - 1.43 1.07, 1.91 Education <12th Grade - - 5.42 3.41, 8.51 High School Grad. - - 2.93 2.01, 4.26 Some College - - 1.41 0.93, 2.12 College Grad. + - - 1.00 ---- Poverty 1979 Not in Poverty - - 1.00 ---- Poverty - - 1.37 1.05, 1.79 å*Model 1 originally included two interaction terms (insurance and race, insurance and SMSA status). The effect of insurance did not differ by race (p=0.6494) or SMSA status (p=0.6543). ** A third model was run with all variables in Model 2 and, in addition, physician population ratio and unemployment rate in the respondent’s present county of residence. None of the least square means for the variables included in Model 2 changed by more than 0.1. Neither unemployment in the respondent’s county of residence at the time of the survey (p = 0.6382) or physician rate was a significant predictor of health status (p=0.3122).
45
Table B-9. Logistic Regression Model Results for Predicting Obesity Model 1* Model 2** O.R. 95% C.I. O.R. 95% CI Insurance coverage ≤ 75% 0.93 0.77, 1.13 0.87 0.71, 1.07 75%-99% 1.05 0.84, 1.30 0.98 0.78, 1.23 100% coverage 1.00 ---- 1.00 ---- Residence >50% in Rural 1.36 1.08, 1.71 1.21 0.95,1.52 >50% in Urban 1.00 ---- 1.00 ---- Race Hispanic 1.51 1.22, 1.85 1.40 1.13, 1.74 Black 1.98 1.64, 2.38 1.84 1.51, 2.25 White 1.00 ---- 1.00 ---- Sex Male - - 0.95 0.80, 1.13 Female - - 1.00 ---- Marital status Never Married - - 1.10 0.87, 1.38 Married - - 1.00 ---- Other - - 0.73 0.59, 0.91 Education <12th Grade - - 1.57 1.10, 2.23 High School Grad. - - 1.66 1.27, 2.16 Some College - - 1.59 1.22, 2.08 College Grad. + - - 1.00 ---- Poverty 1979 Not in Poverty - - 1.00 ---- Poverty - - 1.06 0.85, 1.31 *Model 1 originally included two interaction terms (insurance and race, insurance and SMSA status). The effect of insurance did not differ by race (p=0.1921) or SMSA status (p=0.2175). **A third model was run with all variables including unemployment rate and physician rate per 100,000 people in the respondent’s current county of residence. Neither unemployment rate (p = -/4202) nor physician rate (p = 0.5249) was significant.
46
Table B-10. Logistic Regression Model Results for Predicting Diagnosis of Hypertension Model 1 Model 2** O.R. 95% C.I. O.R. 95% CI Insurance Coverage ≤ 75% 1.15 0.92, 1.45 1.04 0.81, 1.34 75%-99% 1.27 0.99, 1.64 1.19 0.91, 1.56 100% coverage 1.00 ----- 1.00 ----- Residence >50% in Rural 1.07 0.82, 1.39 0.99 0.76, 1.28 >50 % in Urban 1.00 ----- 1.00 ----- Race Hispanic 1.01 0.76, 1.36 0.94 0.69, 1.29 Black 1.79 1.47, 2.19 1.58 1.23, 1.99 White 1.00 ---- 1.00 ---- Sex Male - - 1.06 0.86, 1.30 Female - - 1.00 ---- Marital status Never Married - - 1.18 0.92, 1.51 Married - - 1.00 ---- Other - - 0.97 0.24, 1.19 Education <12th Grade - - 1.47 1.01, 2.14 High School Grad. - - 1.33 1.00, 1.76 Some College - - 1.10 0.78, 1.55 College Grad. + - - 1.00 ---- Poverty 1979 Not in Poverty - - 1.00 ---- Poverty - - 1.07 0.82, 1.54 *Model 1 originally included two interaction terms (insurance and race, insurance and SMSA status). The effect of insurance did not differ by race (p=0.9716) or SMSA status (p=0.3987). **A third model was run with all variables including unemployment rate and physician rate per 100,000 people in the respondent’s current county of residence. Neither unemployment rate (p = 0.7443) nor physician rate (p=0.6645) was significant.
47
Table B-11. Logistic Regression Model Results for Predicting Diagnosis of Diabetes Model 1 Model 2 O.R. 95% C.I. O.R. 95% CI Insurance Coverage ≤ 75% 0.95 0.67, 1.35 0.91 0.55, 1.39 75%-99% 1.16 0.72, 1.85 1.12 0.66, 1.88 100% coverage 1.00 ----- 1.00 ----- Residence >50% in Rural 0.93 0.60, 1.44 0.79 0.49, 1.27 >50 % in Urban 1.00 ----- 1.00 ----- Race Hispanic 2.31 1.51, 3.51 2.22 1.45, 3.41 Black 1.78 1.23, 2.57 1.55 1.00, 2.41 White 1.00 ---- 1.00 ---- Sex Male - - 0.58 0.40, 0.85 Female - - 1.00 ---- Marital status Never Married - - 1.17 0.67, 2.03 Married - - 1.00 ---- Other - - 1.33 0.87, 2.02 Education <12th Grade - - 1.30 0.69, 2.45 High School Grad. - - 1.29 0.76, 2.17 Some College - - 0.81 0.47, 1.41 College Grad. + - - 1.00 ---- Poverty 1979 Poverty - - 0.97 0.65, 1.42 Not in Poverty - - 1.00 ---- *Model 1 originally included two interaction terms (insurance and race, insurance and SMSA status). The effect of insurance did not differ by race (p=0.7195) or SMSA status (p=0.3791). **A third model was run with all variables including unemployment rate and physician rate per 100,000 people in the respondent’s current county of residents. Neither unemployment rate ( p = 0.8628) nor physician rate (p = 0.7998) was significant.
48
Appendix C: References
Abrams J. (2005) Congress Passes Budget That Cuts Medicaid, The Washington Post, 4/29/05, online edition, http://www.washingtonpost.com/wp-dyn/content/article/2005/04/29/AR2005042900234.html. Downloaded May 2, 2005.
Ayanian JZ, Weissman JS, Schneider EC, Ginsburg JA & Zaslavsky, AM. (2000) Unmet health needs of uninsured adults in the United States. JAMA 284:2061-2069.
Baker DW, Sudano JJ, Albert JM, Borawski EA & Dor A (2001) Loss of health insurance and the risk for a decline in self-reported health and physical functioning. New England Journal of Medicine 345 (15): 1106-12.
Baker DW, Sudano JJ, Albert JM, Borawski EA & Dor A (2001) Lack of health insurance and decline in overall health in late middle age. New England Journal of Medicine 345:1106-12.
Baker DW, Sudano JJ, Albert JM, Borawski EA, Dor A. (2002) Loss of health insurance and the risk for a decline in self-reported health and physical functioning. Med Care. 40(11):1126-31.
Baker DW, Sudano JJ. (2005) Health insurance coverage during the years preceding medicare eligibility. Arch Intern Med. Apr 11;165(7):770-6.
Beebe-Dimmer J, Lynch JW, Turrell G, Lustgarten S, Raghunathan T, Kaplan GA.
Childhood and adult socioeconomic conditions and 31-year mortality risk in women. Am J Epidemiol. 2004 Mar 1;159(5):481-90.
Blackwell DL, Hayward MD, Crimmins EM.Does childhood health affect chronic morbidity in later life? Soc Sci Med. 2001 Apr;52(8):1269-84.
Bloomberg LJ and Nichols LM. (2001) The health status of workers who decline employer-sponsored insurance. Health Affairs. 20(6): 180-187.
Bureau of the Census (1981). Characteristics of the Population Below the Poverty Level: 1979. Current Population Reports, Series P-60, No 130. US Government Printing Office, Washington DC.
Comer J, Mueller K, Blankenau J. (2000) Losing and acquiring health insurance: consequences for health care. J Health Soc Policy;11(3):1-15.
Currie J & Gruber J (1996). Health insurance eligibility, utilization of medical care, and child health. Quarterly Journal of Economics 111(2):431-466.
Dansky KH, Dirani R.(1998) The use of health care services by people with diabetes in rural areas. J Rural Health. 14(2):129-37.
Davis, K. (1997) Uninsured in an era of managed care. Health Services Research 31(6):641-649.
Ferrer RL, Palmer R, Burge S. (2005 The family contribution to health status: a population-level estimate. Ann Fam Med. Mar-Apr;3(2):102-8. Franks P, Clancy CM, Gold MR (1993) Health insurance and mortality: Evidence from a national cohort. JAMA 270(6):737-741.
49
Franks P, Clancy CM, Gold MR, Nutting PA (1993) Health insurance and subjective health status: Data from the 1987 National Medical Expenditure Survey. American Journal of Public Health 83(9):1295-9.
Gilmer T, Kronick R. (2005 ) It's The Premiums, Stupid: Projections Of The Uninsured Through 2013. Health Aff (Millwood). Apr 5; [Epub ahead of print]; downloaded April 18, 2005.
Glover S, Moore CG, Samuels ME, Probst JC. (2004) .Disparities in access to care among rural working-age adults. J Rural Health. Summer;20(3):193-205.
Hadley J, Steinberg EP & Feder, J (1991). Comparison of uninsured and privately insured hospital patients: Condition on admission, resource use, and outcome. JAMA 265 (3): 374-379.
Hayward MD, Gorman BK. The long arm of childhood: the influence of early-life social conditions on men's mortality. Demography. 2004 Feb;41(1):87-107
Hoadley JF, Felland LE, Staiti AB. (2004) Federal aid strengthens health care safety net: the strong get stronger. Issue Brief Cent Stud Health Syst Change. Apr;(80):1-4.
Jensen L. (1994) “Employment Hardship and Rural Minorities: Theory, Research, and Policy.” The Review of Black Political Economy 22:125-144.
Kasper JD, Giovannini TA, Hoffman C. (2000) Gaining and Losing Health Insurance: Strengthening the Evidence for Effects on Access to Care and Health Outcomes. Medical Care Research and Review. 57(3):298-318.
Keeler EB, Brook RH, Goldberg GA, Kamberg CJ, Newhouse JP.(1985) How free care reduced hypertension in the health insurance experiment. JAMA;254(14):1926-31.
Kreindel S, Rosetti R, Goldberg R, Savageau J, Yarzebski J, Gore J, Russo A, Bigelow C. (1997) Health insurance coverage and outcome following acute myocardial infarction: A community-wide perspective. Arch Intern Med 157(7):758-62.
Lemak CH, Johnson C, Goodrick EE. (2004) Collaboration to improve services for the uninsured: exploring the concept of health navigators as interorganizational integrators. Health Care Manage Rev. Jul-Sep;29(3):196-206.
Lucas JW, Schiller JS, Benson V. (2004) Summary health statistics for U.S. adults: National Health Interview Survey, 2001.Vital Health Stat 10. Jan;(218):1-134.
Lurie N, Kamberg CJ, Brook RH, Keeler EB, Newhouse JP.(1989) How free care improved vision in the health insurance experiment. Am J Public Health;79(5):640-2.
Manning WG, Newhouse JP, Duan N, Keeler EB, Leibowitz A. (1987) Health Insurance and the Demand for Medical Care: Evidence from a Randomized ExperimentThe American Economic Review, Vol. 77, No. 3. (Jun.) 251-277.
Mensah GA, Mokdad AH, Ford ES, Greenlund KJ, Croft JB. (200 State of disparities in cardiovascular health in the United States. Circulation. Mar 15;111(10):1233-41.
Mueller KJ, Patil K, Ullrich F. (1997) Lengthening spells of uninsurance and their consequences. J Rural Health;13(1):29-37.
New Freedom Commission on Mental Health, Subcommittee on Rural Issues: Background Paper. DHHS. Pub. No. SMA-04-3890. Rockville, MD: 2004.
50
NLSY79 User’s Guide: A Guide to the 1979-2000 National Longitudinal Survey of Youth Data. (2001) Center for Human Resource Research, The Ohio State University, Columbus, Ohio.
Pensola T, Martikainen P. Life-course experiences and mortality by adult social class among young men. Soc Sci Med. 2004 Jun;58(11):2149-70.
Poulton R, Caspi A, Milne BJ, Thomson WM, Taylor A, Sears MR, Moffitt TE.
Association between children's experience of socioeconomic disadvantage and adult health: a life-course study. Lancet. 2002 Nov 23;360(9346):1640-5. Provan KG, Lamb G, Doyle M. (2004) Building legitimacy and the early growth of health networks for the uninsured. Health Care Manage Rev. Apr-Jun;29(2):117-28.
Ross CE & Mirowsky J (2000) Does medical insurance contribute to socioeconomic differences in health? Milbank Quarterly 78(2): 291-320.
Schoen C, DesRoches C. Uninsured and unstably insured: the importance of continuous insurance coverage. Health Serv Res. 2000 Apr;35(1 Pt 2):187-206.
US Preventive Services Task Force (2003) Screening for Obesity in Adults. Downloaded March 8, 2005. http://www.ahcpr.gov/clinic/uspstf/uspsobes.htm
Ware J, Kosinski M, Keller S. 1995. SF-12: How to Score the SF-12 Physical and Mental Health Summary Scales, 2nd edition. Boston: The Health Institute, New England Medical Center.
Wells KB, Manning WG Jr, Valdez RB. (1989) The effects of insurance generosity on the psychological distress and psychological well-being of a general population. Arch Gen Psychiatry; 46(4):315-20.
Wolf DA, Longino CF Jr. (2005) Our "increasingly mobile society"? The curious persistence of a false belief. Gerontologist. Feb;45(1):5-11.
51
Top Related